Visual Question Answering on ScienceVQA
84.8AccViT
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| ViTModel=Idefics2-8B2026.06 | 84.8 | — | — | — | |
| GRIPModel=Idefics2-8B2026.06 | 84.7 | — | — | — | |
| DINOModel=Idefics2-8B2026.06 | 84.6 | — | — | — | |
| GRIPModel=Qwen2.5-VL-7B2026.06 | 84.2 | — | — | — | |
| CLIPModel=Idefics2-8B2026.06 | 83.1 | — | — | — | |
| DINOModel=Qwen2.5-VL-7B2026.06 | 81.9 | — | — | — | |
| CLIPModel=Qwen2.5-VL-7B2026.06 | 81.6 | — | — | — | |
| ViTModel=Qwen2.5-VL-7B2026.06 | 81 | — | — | — | |
| Zero ShotModel=Idefics2-8B2026.06 | 79.2 | — | — | — | |
| Zero ShotModel=Qwen2.5-VL-7B2026.06 | 78.2 | — | — | — | |
| MMBoundaryBase Model=Qwen2VL 7B2025.05 | 71.3 | 34.6 | 42.6 | 56.2 | |
| MMBoundary w/o RECAblation Stage=Reinforcement Learning Stage2025.05 | 71.2 | 36.3 | 42.6 | 54.8 | |
| MMBoundary2025.05 | 70.3 | 35.4 | 39.2 | 56.5 | |
| Conf-CSRBase Model=Qwen2VL 7B2025.05 | 70.2 | 43.3 | 53.4 | 51.7 | |
| MMBoundary w/o ULNLPAblation Stage=Warm-Up Stage2025.05 | 69.8 | 36.9 | 42.1 | 54.8 | |
| Conf-CSR2025.05 | 69.4 | 45.3 | 50.3 | 50.2 | |
| MMBoundary w/o UTSARAblation Stage=Warm-Up Stage2025.05 | 69.4 | 35.2 | 41.7 | 54.6 | |
| MMBoundary w/o RCSAblation Stage=Reinforcement Learning Stage2025.05 | 69.3 | 37.2 | 44.9 | 55.6 | |
| RCEBase Model=Qwen2VL 7B2025.05 | 68.6 | 40.1 | 48.3 | 50.4 | |
| MMBoundary w/o RKAAblation Stage=Reinforcement Learning Stage2025.05 | 68.6 | 33.4 | 41 | 53.5 | |
| MMBoundary w/o S-SMappingAblation Stage=Warm-Up Stage2025.05 | 68.4 | 37.7 | 44.3 | 53.1 | |
| MMBoundary w/o UMTEAblation Stage=Warm-Up Stage2025.05 | 68.2 | 38.5 | 44.1 | 53.6 | |
| MMBoundary w/o UCLIPSAblation Stage=Warm-Up Stage2025.05 | 68.1 | 37.2 | 43.5 | 53.2 | |
| RCE2025.05 | 67.1 | 41.3 | 47.5 | 49.7 | |
| MMBoundary w UMaxAblation Stage=Warm-Up Stage2025.05 | 66.3 | 39.1 | 46.7 | 49.4 | |
| MMBoundary w/o RLAblation Stage=Reinforcement Learning Stage2025.05 | 66.3 | 41.9 | 48.1 | 49.5 | |
| DRL2025.05 | 65.4 | 47.6 | 51.3 | 48.5 | |
| DRLBase Model=Qwen2VL 7B2025.05 | 64.1 | 43.5 | 52.3 | 48.7 | |
| CE-EnsembleDistance Metric=Cosine Distance2025.04 | 63.7 | — | — | — | |
| SaySelf2025.05 | 63.3 | 38.6 | 46.2 | 48.3 | |
| SaySelfBase Model=Qwen2VL 7B2025.05 | 62.8 | 33.2 | 44.5 | 52.3 | |
| MultisampleBase Model=Qwen2VL 7B2025.05 | 61.3 | 46.7 | 49.2 | 41.5 | |
| Baseline2025.04 | 61.3 | — | — | — | |
| SC2025.05 | 60.2 | 46.3 | 53.4 | 44.2 | |
| Multisample2025.05 | 59.6 | 44.6 | 50 | 43.4 | |
| DPV2025.05 | 58.2 | 59.3 | 61.1 | 41.4 | |
| DPS2025.05 | 57.5 | 57.4 | 57.5 | 42.7 | |
| Baseline2025.10 | — | 28.09 | — | — | |
| CutMixVision Token Count=Full, Training=ranking loss2025.10 | — | 24.71 | — | — | |
| CutMixVision Token Count=50%, Training=ranking loss2025.10 | — | 20.63 | — | — | |
| MergeMixVision Token Count=Full, Training=ranking loss2025.10 | — | 23.66 | — | — | |
| MergeMixVision Token Count=50%, Training=ranking loss2025.10 | — | 25.22 | — | — | |
| MixUpVision Token Count=Full, Training=ranking loss2025.10 | — | 21.51 | — | — | |
| MixUpVision Token Count=50%, Training=ranking loss2025.10 | — | 22.61 | — | — | |
| ResizeMixVision Token Count=Full, Training=ranking loss2025.10 | — | 24.58 | — | — | |
| ResizeMixVision Token Count=50%, Training=ranking loss2025.10 | — | 23.74 | — | — | |
| SFT VisionVision Token Count=Full2025.10 | — | 21.66 | — | — | |
| SFT VisionVision Token Count=50%2025.10 | — | 24.28 | — | — |